Skip to content

sklearn.md

Latest commit

 

History

History
151 lines (126 loc) · 6.86 KB

sklearn.md

File metadata and controls

151 lines (126 loc) · 6.86 KB

scikit-learn (sklearn)

Pure NumPy reimplementation | v1.0.0-offlinai | 12,000+ lines | 40 modules

Runs 100% offline on iPad. No compiled C extensions — everything is pure Python + NumPy.

What's Included

Classification & Regression

Module Classes Notes
linear_model LinearRegression, Ridge, Lasso, ElasticNet, SGDClassifier, SGDRegressor, LogisticRegression, Perceptron, PassiveAggressiveClassifier, BayesianRidge, ARDRegression, HuberRegressor, RANSACRegressor, TheilSenRegressor, QuantileRegressor, OrthogonalMatchingPursuit, Lars, LassoLars 18 classes
tree DecisionTreeClassifier, DecisionTreeRegressor Gini & variance splitting
neighbors KNeighborsClassifier, KNeighborsRegressor Brute-force, uniform/distance weights
svm SVC, SVR, LinearSVC, LinearSVR, NuSVC, NuSVR, OneClassSVM SGD-based linear SVM
naive_bayes GaussianNB, MultinomialNB, BernoulliNB, ComplementNB, CategoricalNB 5 variants
neural_network MLPClassifier, MLPRegressor Multi-layer perceptron with backprop

Ensemble Methods

Class Description
RandomForestClassifier / Regressor Bagged decision trees
GradientBoostingClassifier / Regressor Gradient boosted trees
HistGradientBoostingClassifier / Regressor Histogram-based gradient boosting
AdaBoostClassifier / Regressor Adaptive boosting
ExtraTreesClassifier / Regressor Extremely randomized trees
BaggingClassifier / Regressor Bootstrap aggregating
VotingClassifier / Regressor Soft/hard voting ensemble
StackingClassifier / Regressor Stacked generalization
IsolationForest Anomaly detection via random forests

Clustering

Class Description
KMeans, MiniBatchKMeans K-Means with k-means++ init
DBSCAN Density-based spatial clustering
AgglomerativeClustering Hierarchical (single, complete, average, ward)
SpectralClustering Graph-based spectral method
MeanShift Mode-seeking clustering
OPTICS Ordering points for cluster structure
Birch Balanced iterative reducing
AffinityPropagation Message-passing clustering
HDBSCAN Hierarchical DBSCAN
GaussianMixture (in mixture) EM-based Gaussian mixture models
BayesianGaussianMixture Variational inference GMM
FeatureAgglomeration Feature-space hierarchical clustering

Preprocessing

Class Description
StandardScaler Zero-mean, unit-variance
MinMaxScaler Scale to [0, 1]
MaxAbsScaler Scale by max absolute value
RobustScaler Median/IQR scaling
Normalizer L1/L2 row normalization
Binarizer Threshold binarization
LabelEncoder String-to-int labels
OneHotEncoder One-hot encoding
OrdinalEncoder Ordinal encoding
LabelBinarizer Multi-class binarization
MultiLabelBinarizer Multi-label binarization
PolynomialFeatures Polynomial cross-features
FunctionTransformer Custom transform function
PowerTransformer Yeo-Johnson / Box-Cox
QuantileTransformer Quantile normalization
KBinsDiscretizer Binning continuous features
SplineTransformer B-spline feature expansion
TargetEncoder Target-based encoding

Decomposition & Manifold

Class Description
PCA Principal Component Analysis
TruncatedSVD SVD for sparse-like data
KernelPCA Kernel-based PCA
IncrementalPCA Batch-wise PCA
FastICA Independent Component Analysis
NMF Non-negative Matrix Factorization
FactorAnalysis Factor analysis
LatentDirichletAllocation Topic modeling
DictionaryLearning Sparse coding
SparsePCA Sparse PCA
MiniBatchDictionaryLearning Online dictionary learning
TSNE (in manifold) t-distributed SNE
Isomap Isometric mapping
MDS Multidimensional scaling
LocallyLinearEmbedding LLE
SpectralEmbedding Spectral embedding

Model Selection

Item Description
train_test_split() Split with stratification
cross_val_score() K-fold cross validation
GridSearchCV Exhaustive grid search
RandomizedSearchCV Random hyperparameter search
KFold / StratifiedKFold K-fold splitters
RepeatedKFold Repeated K-fold
LeaveOneOut / LeavePOut LOO / LPO
ShuffleSplit / StratifiedShuffleSplit Random splits
GroupKFold / GroupShuffleSplit Group-aware splits
TimeSeriesSplit Time-ordered splits
ParameterGrid / ParameterSampler Parameter iteration
learning_curve / validation_curve Diagnostic curves
cross_validate Multi-metric CV

Metrics (38 functions)

Classification: accuracy_score, precision_score, recall_score, f1_score, roc_auc_score, log_loss, confusion_matrix, classification_report, matthews_corrcoef, cohen_kappa_score, balanced_accuracy_score, top_k_accuracy_score, average_precision_score, brier_score_loss, hamming_loss, hinge_loss, jaccard_score, zero_one_loss, roc_curve, precision_recall_curve

Regression: mean_squared_error, mean_absolute_error, r2_score, mean_squared_log_error, median_absolute_error, explained_variance_score, max_error, mean_absolute_percentage_error, mean_pinball_loss, d2_pinball_score

Clustering: silhouette_score, silhouette_samples, calinski_harabasz_score, davies_bouldin_score, adjusted_rand_score, adjusted_mutual_info_score, normalized_mutual_info_score, homogeneity_completeness_v_measure

Pipeline & Datasets

Item Description
Pipeline Chain transformers + estimator
make_pipeline() Auto-named pipeline
ColumnTransformer Per-column transformers
FeatureUnion Parallel feature pipelines
make_classification() Synthetic classification data
make_regression() Synthetic regression data
make_blobs() Gaussian blob clusters
make_moons() / make_circles() Non-linear datasets
load_iris() 150 samples, 4 features, 3 classes
load_digits() 1,797 samples, 64 features, 10 classes
load_wine() 178 samples, 13 features, 3 classes
load_breast_cancer() 569 samples, 30 features, 2 classes
load_diabetes() 442 samples, 10 features
load_boston() 506 samples, 13 features
load_linnerud() 20 samples, physiological data
fetch_california_housing() 20,640 samples, 8 features

Limitations

  • Linear SVM only (no RBF/polynomial kernels requiring QP solver)
  • No sparse matrix support (no scipy.sparse)
  • Manifold methods may be slow on large datasets
  • Neural network limited to standard MLP architecture